v1.16.0

Minor Changes

• 1333721: Add ContentScanner for SKILL.md security scanning

  Changes:

  • New ContentScanner module with 6 detection rules: prompt injection, data exfiltration, content obfuscation, sensitive file access, stealth instructions, and oversized content
  • Context-aware scanning: skips safe zones (frontmatter, code blocks, blockquotes, inline code, quoted text) to reduce false positives
  • Configurable via ScannerOptions: override rule levels, disable rules, add custom rules
  • Integrated into reskill publish — high-risk content blocks publishing, medium-risk shows warnings
  • --dry-run also runs content scan for author self-checking
  • New subpath export reskill/scanner for lightweight server-side usage

  ---

  新增 ContentScanner，用于 SKILL.md 内容安全扫描

  变更：

  • 新增 ContentScanner 模块，包含 6 类检测规则：prompt injection、数据泄露、内容混淆、敏感文件访问、隐蔽指令、超大内容
  • 上下文感知扫描：自动跳过安全区域（frontmatter、代码块、引用、行内代码、引号内文本），降低误报率
  • 支持通过 ScannerOptions 配置：覆盖规则等级、禁用规则、添加自定义规则
  • 集成到 reskill publish：高风险内容阻止发布，中风险显示警告
  • --dry-run 模式也执行扫描，方便作者自检
  • 新增子路径导出 reskill/scanner，供服务端轻量引入

Patch Changes

• 179193f: Add Chinese pattern detection for prompt-injection and stealth-instructions rules

  Changes:

  • Add 11 Chinese regex patterns to prompt-injection rule (high risk)
  • Add 6 Chinese patterns + 1 safe exclusion to stealth-instructions rule (medium risk)
  • Fix multilingual bypass vulnerability where Chinese translations of malicious content bypassed all detection

  ---

  为 prompt-injection 和 stealth-instructions 规则添加中文模式检测

  变更：

  • prompt-injection 规则新增 11 条中文正则匹配（高风险）
  • stealth-instructions 规则新增 6 条中文匹配 + 1 条安全排除（中风险）
  • 修复多语言绕过漏洞：中文翻译的恶意内容可完全绕过检测